Interface ChatMistralAIInput

Input to chat model class.

interface ChatMistralAIInput {
    apiKey?: string;
    endpoint?: string;
    maxTokens?: number;
    modelName?: string;
    randomSeed?: number;
    safeMode?: boolean;
    safePrompt?: boolean;
    streaming?: boolean;
    temperature?: number;
    topP?: number;
}

Hierarchy (view full)

Implemented by

Properties

apiKey? endpoint? maxTokens? modelName? randomSeed? safeMode? safePrompt? streaming? temperature? topP?

Properties

Optional apiKey

apiKey?: string

The API key to use.

Default

{process.env.MISTRAL_API_KEY}

Optional endpoint

endpoint?: string

Override the default endpoint.

Optional maxTokens

maxTokens?: number

The maximum number of tokens to generate in the completion. The token count of your prompt plus max_tokens cannot exceed the model's context length.

Optional modelName

modelName?: string

The name of the model to use.

Default

{"mistral-small-latest"}

Optional randomSeed

randomSeed?: number

The seed to use for random sampling. If set, different calls will generate deterministic results.

Optional safeMode

safeMode?: boolean

Whether to inject a safety prompt before all conversations.

Default

{false}

⚠️ Deprecated ⚠️

use safePrompt instead

This feature is deprecated and will be removed in the future.

It is not recommended for use.

Optional safePrompt

safePrompt?: boolean

Whether to inject a safety prompt before all conversations.

Default

{false}

Optional streaming

streaming?: boolean

Whether or not to stream the response.

Default

{false}

Optional temperature

temperature?: number

What sampling temperature to use, between 0.0 and 2.0. Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic.

Default

{0.7}

Optional topP

topP?: number

Nucleus sampling, where the model considers the results of the tokens with top_p probability mass. So 0.1 means only the tokens comprising the top 10% probability mass are considered. Should be between 0 and 1.

Default

{1}

Settings

Member Visibility

Theme

On This Page

Generated using TypeDoc